Date of Award

Summer 2004

Document Type

Thesis - Restricted

Degree Name

Master of Science (MS)


Electrical and Computer Engineering


Neural networks have been applied in QTL analysis because of tbeir ability to access all genetic markers simultaneously. In the neural network genetic analysis methods, genotype data is presented to the network as input and phenotype data as output. After the training phase, the network weights are analyzed to obtain a contribution measure for each marker locus to the trait. In this thesis work, a more comprehensive sensitivity measure for evaluating the impact of each marker locus on the trail value is developed. This sensitivity analysis approach is tested on the backcross generation and compared with the LOD score in the conventional interval mapping. The significance of these sensitivities is also estimated by using the permutation test. In this research, we also established a preliminary framework for linkage analysis using internal parameters of a properly trained neural network. The mathematical expressions of the weight and bias with respect to the recombination fraction are given for various neuron and network models, and a method to estimate the true QTL location between two flanking markers by using the ratio of weights is demonstrated. These results are helpful to integrate the neural network model more deeply into linkage analysis.



Restricted Access Item

Having trouble?